Skip to content

feat(api): accept a caller-supplied bbox to skip YOLO detection#46

Draft
MateoLostanlen wants to merge 3 commits into
mainfrom
feat/api-forced-bbox
Draft

feat(api): accept a caller-supplied bbox to skip YOLO detection#46
MateoLostanlen wants to merge 3 commits into
mainfrom
feat/api-forced-bbox

Conversation

@MateoLostanlen

Copy link
Copy Markdown
Member
  • The alert platform already runs a YOLO detector and holds the smoke bboxes when it calls /predict; today the API re-runs its own YOLO over every frame, computing the same thing twice. This adds optional bbox_xyxyn / bbox_confidence to POST /predict: the supplied box is taken as the detection for every frame, building one full-length tube and skipping the detector entirely for faster inference.
  • Rides the existing predict(frame_detections=...) injection seam via a new pure make_forced_detections() core helper; no second inference path. The context margin (context_factor) is still applied server-side, so callers pass the tight detection box.
  • Mutually exclusive with roi_xyxyn (400); the detection cache is bypassed so synthetic boxes never pollute it.
  • Spec in docs/specs/2026-06-11-api-forced-bbox-design.md.

@Chouffe

Chouffe commented Jun 11, 2026

Copy link
Copy Markdown
Collaborator

Implementation in #47

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants